Application of multilayer perceptron to deep reinforcement learning for stock market trading and analysis

Trading strategies to maximize profits by tracking and responding dynamic stock market variations is a complex task. This paper proposes use multilayer perceptron method (a part of artificial neural networks (ANNs)), that can be used deploy deep reinforcement learn the process predicting analyzing products with aim profit making. We trained agent using four algorithms: proximal policy optimization (PPO), Q-learning (DQN), deterministic gradient (DDPG) method, advantage actor critic (A2C). The proposed system, comprising these algorithms, tested real time data two products: Dow Jones (DJIA-index), Qualcomm (shares). performance linked individual algorithms was evaluated, compared analyzed Sharpe ratio, Sortino Skew Kurtosis, thus leading most effective algorithm being chosen. Based on parameter values, maximizes making for respective financial product determined. also extended same approach study ascertain predictive trading under highly volatile scenario, such as pandemic coronavirus disease 2019 (COVID-19).